The Model Context Protocol, or MCP, is an open standard for connecting AI applications to external tools and data sources. A tool or data source exposes its functions through an MCP server, and an AI application can connect to it and use those functions, without needing a custom integration for each one.
It matters because it makes it easier to give AI assistants access to files, databases and services. It also matters for security, because an MCP server often needs credentials to act on a user's behalf, and those credentials are commonly kept in a local configuration file where they can be exposed.
The standard does not decide what is safe to connect. A server can be given broad access, and one that is poorly built or malicious can create risks such as exposed credentials or instructions hidden in the data it returns. Support and details differ between applications and change as the standard develops.
To use it carefully, connect only servers you trust, give each the narrowest credentials it needs, keep those credentials out of shared files and review what each server can do.